Data Readiness: The Unglamorous Prerequisite

When a mid-market AI project fails, the post-mortem usually blames the model, the vendor, or the ambition. The real cause is more often the least glamorous part of the stack: the data. Most AI failures are data failures wearing an AI costume.

This is unintuitive, because the demo worked. The demo always works — it runs on a clean, curated sample the vendor prepared. Production runs on your data: the version with duplicate records, blank fields, three spellings of the same client name, and a definition of "active customer" that differs between two departments. AI does not fix that. It amplifies it.

What "data ready" actually means

Readiness is not having a lot of data. It is having data the system can use. Four properties matter:

• Accessible. Can the data be reached without a three-week extract from a system no one fully maintains? If it's locked in a format or a silo, it isn't usable yet.
• Accurate. Is it correct and current enough to act on? A model trained on stale or wrong data produces confident, wrong answers — the most dangerous kind.
• Governed. Do you know what it contains, who may use it, and under which obligations (PDPA, Privacy Act, PCPD)? Ungoverned data is a compliance incident waiting for a trigger.
• Connected. Does it join up across systems? Value usually lives in the relationships between datasets, not in any one of them alone.

Why this is the work no one wants

Data remediation is unglamorous, hard to demo, and easy to defer. It produces no screenshot for the board deck. So it gets skipped — and the project proceeds on the assumption that the data is fine, until week six of implementation reveals that it isn't. By then the spend is committed and the timeline is public.

The firms that succeed treat data readiness as the first phase, not a surprise in the middle. They scope the remediation honestly, sequence it ahead of the build, and accept that the unglamorous work is what makes the glamorous part possible.

You can assess it before you commit

The good news is that data readiness is knowable in advance. A structured assessment surfaces the gaps — fragmentation, quality, governance, integration — before a build is funded against assumptions that won't hold. That is precisely the kind of finding that changes a project's sequence and saves its budget.

The Prime Diagnostic™ assesses data foundations as one of the five PRIME pillars — Resource Architecture — so you know what to fix before implementation, not during it.